Cash flow vs. collateralbased credit - CORE

Cash flow vs. collateral-basedcreditPerformance of micro, small and medium-sizedfirms in transition economies1

Francesca Cassano, Karin J~oeveer* and Jan Svejnar***London School of Economics. E-mail: [email protected]

**Columbia University, CERGE-EI, CEPR, IZA. E-mail: [email protected]

Abstract

We study factors affecting micro, small and medium-sized enterprises (MSMEs)receiving loans and the effect of these loans on MSMEs performance. We study twotypes of loans – a new type based on cash flows and a traditional-style loan basedon collateral. We use unique surveys of MSMEs from Bulgaria, Georgia, Russia andUkraine. We find that MSMEs receiving a cash flow or collateral loan in the past aremore likely to receive the same type of loan (and larger sized) in the future and thatcash flow loans may be the preferred form of credit. Both types of loans are relatedpositively to most performance indicators, enabling the MSMEs for instance to bemore profitable and expand production. The cash flow loans also appear to be par-ticularly attractive credit delivery schemes for micro and small enterprises. Finally,the effects of the smallest loans are often negative, suggesting that the minimumloan size is an important policy issue.

JEL classification: G21, G31, O16.Keywords: Micro, small and medium-sized enterprises, bank credit, firm perfor-mance, emerging markets.

Received: December 11, 2009; Acceptance: December 6, 2012

1 The article was written with the financial and institutional support of the Japan Europe Development Fundand EBRD. We thank members of the Office of the Chief Economist at EBRD, Leora Klapper and participantsat various seminars for useful comments. Jan Svejnar also benefitted from a grant of the Grant Agency of theCzech Republic (Grant P402/10/2130). Francesca Cassano was formerly known as Francesca Pissarides. Theusual disclaimer applies.

Economics of TransitionVolume 21(2) 2013, 269–300DOI: 10.1111/ecot.12016

� 2013 The AuthorsEconomics of Transition � 2013 The European Bank for Reconstruction and Development.Published by Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main St, Malden, MA 02148, USA

brought to you by COREView metadata, citation and similar papers at core.ac.uk

provided by Columbia University Academic Commons

https://core.ac.uk/display/161454657?utm_source=pdf&utm_medium=banner&utm_campaign=pdf-decoration-v1

1. Introduction

Due to the important contribution that entrepreneurs and micro, small and med-ium-sized enterprises (MSMEs) make to economic growth, innovation and employ-ment creation, researchers and policymakers emphasize the need for a betterunderstanding of the factors influencing the rise and performance of these firms.Academic research has identified limitations to the availability of finance for MSMEsconstraining firms’ performance, such as informational asymmetries between bor-rowers and lenders, lack of credit history on the part of many MSMEs, poor legaland institutional infrastructure, scarcity of appropriate credit skills in banks andeconomies of scale in lending. In this article, we analyse the effects of the provisionof two types of bank credit to MSMEs in transition economies on firm performance –a new type of loan based on cash flow and traditional-style loans based on collateral(fixed assets and movable assets such as cars).

Before the 2001–2004 period covered by our study, standard bank credit toMSMEs in the transition economies was very limited, with microenterprises obtain-ing virtually no credit. The few firms that benefited from bank credit had to providelarge amounts of collateral and sometimes had to use a series of short-term loans tofinance longer-term capital investments.2 In most cases, the type of collateralaccepted by financial institutions was not available to MSMEs. There was hence animportant gap in the financial market.3

To overcome the problem of a lack of collateral on the part of MSMEs, govern-ments, international financial institutions and non-governmental organizations(NGOs) established new programmes to support the delivery of (cash flow based)credit to MSMEs. The rationale was to support the creation of sustainable and com-mercially viable microfinance channels. These programmes focused on reducinglending costs, lowering banks’ perceptions of risk associated with MSME borrow-ers, improving banks’ screening methodology and helping these borrowers build acredit history. The ultimate objective was the easing of credit constraints ofMSMEs.

Interestingly, while the objectives of MSME lending programmes are widelyaccepted as being important, little evidence is available regarding their impact (seeBrown et al., 2002; Hulme and Mosley, 1996; Morduch, 1999). Most evaluations havebeen monitoring exercises relying on the perceptions of the beneficiaries of the pro-grammes. Those evaluations do not satisfactorily address the issue of selection bias,that is, the problem that the observed performance of the beneficiaries may not beattributable solely to the programme, but also to predetermined characteristics ofthe firms that allowed them to be selected into the programmes. Since inherently

2 Using survey data from Kosovo, Krasniqi (2010) evaluates the determinants of small firms receiving a loan.He shows that collateral is the main indicator on the basis of which banks make their lending decisions.3 See for instance Pissarides, Singer and Svejnar (2003) for an analysis of the objectives and constraints ofentrepreneurs in transition economies.

� 2013 The AuthorsEconomics of Transition � 2013 The European Bank for Reconstruction and Development

270 Cassano , Joeveer and Svejnar

better performing MSMEs are more likely to be selected into the programmes, theimpact of programmes is normally over-estimated.4

More recent evaluations strive to eliminate the selection bias by using statisticaltechniques such as difference-in-differences, difference-in-differences-in-differencesand instrumental variables (IVs).5 In some cases, researchers have used other quasi-experimental techniques,6 such as trying to assess the impact of microfinance pro-grammes by comparing the impact on repeat clients of a microfinance programmeto that on new clients, where new clients are the control group and repeat clientsconstitute the treatment group.7 This methodology has potentially serious shortcom-ings, however, in that it omits dropouts from the analysis (see, for example, discus-sion in Karlan, 2001).8 Alternatively, Coleman (1999) utilized pipeline matching,whereby borrowers’ performance is compared to that of clients who sign up to par-ticipate in a future lending programme and thus undergo the same selection as cur-rent borrowers. Although appealing, this methodology may yield a biased estimateof the true impact in the presence of the Ashenfelter dip effect (in other words, thepre-programme performance of the control group may be affected by the expecta-tion of being in the programme).9 Bah et al. (2011) evaluate the effect of technicaland financial aid in Macedonia. They use matching techniques to address the selec-tion bias and find that the assistance improves firms’ employment growth.

In this article, we are fortunate to have obtained access to client data from banksparticipating in microfinance programmes of the European Bank for Reconstructionand Development (the EBRD) for an ex post analysis.10 The EBRD is among thekey institutions introducing and supporting the commercially based credit

4 See, for example, Mosley (1998).5 See, for example, Banerjee and Duflo (2004); LeLarge et al. (2008); Storey (2000). These studies control for theeffect of observed variables that affect the outcome and may be correlated with participation in the pro-gramme by including these variables as regressors in the estimation equation predicting the outcome, match-ing beneficiaries and non-beneficiaries that are similar in observed characteristics, or modelling how selectioninto a programme occurs and how it is correlated with unobserved variables. Other analytical evaluations userandomly selected groups of beneficiaries and non-beneficiaries. Gin�e et al. (2006) for instance discuss arecently launched randomized impact evaluation of a microfinance programme in the Philippines.6 Karlan and Goldberg (2006) provide a useful survey of studies examining the impact of microfinance pro-grammes, products and policies carried out with the randomized trials methodology and quasi-experimentaldesigns.7 This approach has been used in impact evaluations funded by USAID through its AIMS project.8 Alexander-Tedeschi and Karlan (2009) for instance find that by including dropouts in the analysis the esti-mated impact of microfinance is greatly reduced. They compare the different results of the impact evaluationof a microfinance programme in Peru with and without dropouts in the sample. The analysis excluding drop-outs found a positive impact of credit on profits, household income and firm employment. The analysisincluding dropouts found a negative impact of credit on firm profits, and a reduced positive impact on house-hold income and firm employment.9 Coleman (1999) found no impact of credit on the performance of clients and prospective clients of a microfi-nance programme in Thailand.10 At the time, a randomized ex ante evaluation approach was not feasible because the financial intermediariesperceived the cost of randomized provision of credit to be too high relative to the potential benefits.


Cash Flow vs. Collateral-based Credit 271

programmes targeting MSMEs in the Eastern European region. The EBRD intro-duced cash flow-based lending that uses a more flexible definition of collateral,therefore reaching out to a larger number of firms. The EBRD programme providedfinancial incentives to banks (in the form of credit lines granted at a discounted rate)to use this new cash flow-based approach in lending to small borrowers and it alsoaimed at building credit skills for MSME lending in existing commercial banks andnewly established specialized banks known as microfinance banks.

To carry out our analysis, we administered a survey in 2005 to a sample of (a)MSMEs that had received a loan from one of eight EBRD-sponsored MSME lendingprojects in 2002, and (b) similar MSMEs that had never received an EBRD projectloan. The latter sample represents our control group. In both groups, some firmshad received loans from non-EBRD sources during our sample period and somehad not. In the survey, we have data on performance indicators and all loans (bothcash flow and collateral loans) that the firms obtained during 2001–2004.

There are two key questions that we address. First, what factors determinewhether MSMEs receive collateral loans or cash flow loans? Second, what is theeffect of the cash flow vs. collateral credit on MSME performance? We use severalindicators of firm performance: capital (fixed assets) formation, revenues, employ-ment and profit.11 We conduct our analyses on all firms taken together, as well ason specific size groups: micro, small and medium-sized enterprises, defined as firmswith 1–5 employees (including the self-employed and working family members),6–15 employees and 16 or more employees, respectively.

We find that firms that received a given (cash flow or collateral) loan in the pastare more likely to receive the same type of loan, and also a larger loan of the sametype, in the future. Having received a cash flow loan in the past has a negative effecton the probability of receiving a collateral loan in the future, while the correspondingnegative cross-effect from receiving collateral loans in the past on the probability ofreceiving a cash flow loan is statistically insignificant for micro and small firms. Esti-mates based on the entire sample suggest that both the cash flow and collateral loanshave a positive relationship with fixed asset formation, suggesting that firms usebank loans for investment in fixed capital. In terms of dollars of fixed assets generatedby a dollar of loan, the effects of the two types of loans are similar. The positive effectof both types of loans is by and large also found with respect to revenues, andemployment. In particular, the overall estimates for all firms indicate that the loansserve the purpose of enabling theMSMEs to expand production beyond the scale thatthey could have achieved without this source of credit. In the overall sample, the twotypes of loans are also found to have a positive effect on profitability. We find that theabove pattern holds across the size groups of firms, but some estimated effects (espe-cially those for micro firms and to a lesser extent medium-sized firms) are statistically

11 Berger and Udell’s (2006) work highlights the fact that programmes providing credit through differentlending methodologies are likely to yield different outcomes and should not be treated as a homogeneous setof providers.



less significant. Finally, our estimates suggest that while most cash flow and collat-eral loans have a positive effect on the growth of fixed assets, revenues, employmentand net profit, in many cases the effects of the smallest loans are negative.

The article is organized as follows. Section 2 describes the MSME financing pro-jects whose impact we evaluate, while Section 3 discusses the main features of thesurvey and basic statistics. Section 4 outlines the analytical framework, Section 5presents the empirical results and Section 6 contains the conclusions. Appendix 1provides descriptive statistics on the treatment and control firms across size groups,while Appendix 2 presents information on the banks that provided cash flow credit.

2. Objectives and structure of the evaluated MSME projects

The EBRD financing projects are structured as loans to banks and other financial insti-tutions that use the EBRD loans to extend loans to MSME borrowers. The EBRD cou-ples these loans to banks with a technical assistance programme through which thebanks acquire new lending methodologies that are appropriate for dealing with alarge number of small borrowers in an environment with underdeveloped institu-tional andfinancial infrastructure. The EBRDprojects result in individualMSME loansthat are extended on the basis of risk considerations and borrowers’ ability to repaythe loan (measured as a percentage of their cash flow projections).12 Banks not partici-pating in the EBRD programmes have instead been extending loans to small borrow-ers using a traditional collateral lendingmethodology. The cash flow approach allowslocal banks to extend a large number of loans and in a short loan-processing time.From Table 1, we see that the number of days to receive a loan is statistically signifi-cantly lower for cash flow loans than for collateral loans. The short processing time inturn permits MSMEs to access credit when they need it. The EBRD programmes pro-vide no subsidy to the MSMEs. The effective interest rates charged by the EBRD pro-grammes are on average in line with the market rates charged by other banks. FromTable 1, we can see that cash flow loans in fact charge slightly higher interest rates. Interms of collateral requirements, however, the banks in the EBRD programmes acceptas collateral almost anything that ‘matters to the borrower’ to provide the borrowerwith an incentive to repay the loan, but not to protect itself in case of default. The useof collateral in these loans is in fact purely for incentive purposes because the net reve-nue that the bankmight obtain from the sale of such flexible collateral would be negli-gible but the potential loss to the borrower would often exceed the value of the loan.FromTable 1,we see that themean percentage of loan provided as collateralwas simi-lar for the two types of loans in 2001,while in 2004 the treatment firmswere pledging ahigher share of collateral for both types of loans comparedwith control firms.

The EBRD operates through de novo dedicated microfinance banks as well asexisting local commercial banks. The microfinance banks are set up by both private

12 The main variable of interest is the leverage ratio of the borrower.



Tab

le1.

Loa

nstatistics

Con

trol

grou

pTreatmen

tgroup

Collateral

loan

sCollateral

loan

sCashflow

loan

sDifference

betwee

n1an

d2

T-test

Difference

between

2an

d3

T-test

Difference

betwee

n1an

d3

T-test

12

34

56

78

9

2001 Meanloan

size

(inUSD

s)

58,039

.10

(99,08

6.7)

11,002

.60

(12,20

1.39

)13

,737

.59

(28,31

7.67

)47

,036

.50

3.25

***

�2,734

.99

�0.99

44,301

.51

3.07

***

Meanloan

maturity

(inmon

ths)

15.31

(11.31

)20

.03

(16.59

)12

.73

(4.21)

�4.72

�1.44

7.30

2.55

**2.58

1.56

Meanloan

interest

rate

(in%)

19.23

(5.54)

17.67

(6.23)

20.97

(7.08)

1.56

1.17

�3.30

�2.85***

�1.74

�1.9*

Mean

percen

tage

ofloan

prov

ided

ascolla

teral

177.71

(116

.79)

176.03

(96.48

)21

8.85

(502

.92)

1.68

0.07

�42.82

�1.20

�41.14

�1.15

Mean

numbe

rof

working

day

stake

nto

obtain

the

loan

20.96

(24.67

)25

.12

(19.94

)13

.99

(16.46

)� 4

.16

�0.84

11.13

3.11

***

6.97

1.89

*



Tab

le1.

(Con

tinued

)

Con

trol

grou

pTreatmen

tgroup

Collateral

loan

sCollateral

loan

sCashflow

loan

sDifference

between

1an

d2

T-test

Difference

betwee

n2an

d3

T-test

Difference

betwee

n1an

d3

T-test

12

34

56

78

9

Num

berof

loan

s48

3425

2

2004 Meanloan

size

(inUSD

s)

40,997

.91

(87,531.86)

35,079

.05

(60,22

0.4)

22,185

.65

(46,00

6.73

)5,91

8.86

0.48

12,893

.40

1.47

18,812

.25

2.04

**

Meanloan

maturity

(inmon

ths)

21.58

(21.19

)27

.27

(24.28

)19

.07

(13.04

)�5

.69

�1.40

8.20

2.35

**2.51

1.11

Meanloan

interestrate

(in%)

17.11

(5.76)

16.25

(4.72)

18.27

(6.68)

0.86

0.96

�2.02

�2.76***

�1.17

�1.76*

Mean

percen

tage

ofloan

prov

ided

ascolla

teral

137.52

(70.34

)17

9.50

(98.97

)16

5.34

(118

.46)

�41.98

�2.67***

14.16

0.94

�27.83

�3.09***

Meannu

mbe

rof

working

day

stake

nto

obtain

theloan

16.09

(17.77

)17

.08

(19.78

)9.39

(10.62

)�0

.99

�0.30

7.69

2.71

***

6.70

3.55

***



Tab

le1.

(Con

tinued

)

Con

trol

grou

pTreatmen

tgroup

Collateral

loan

sCollateral

loan

sCashflow

loan

sDifference

betwee

n1an

d2

T-test

Difference

betwee

n2an

d3

T-test

Difference

betwee

n1an

d3

T-test

12

34

56

78

9

Num

berof

loan

s95

5048

1

Notes:T

reatmen

tgrou

pconsists

ofMSM

Esthat

hadreceived

aloan

from

oneof

eigh

tEBRD-spo

nsored

MSM

Elend

ingprojects

in2002.C

ontrol

grou

pconsistsof

MSM

Esthat

hadne

verreceived

anEBRD

projectloa

n.Loa

nsize

isad

justed

toprod

ucer

prices.F

orallcou

ntries

thecoun

trysp

e-cificprod

ucer

priceindex

isus

edexcept

forRus

siaforwhich

theNizhn

yNov

gorodregion

alprod

ucer

priceinde

xisus

ed.S

tand

arddev

iatio

nsare

give

nin

parenthe

ses.*sign

ificant

at10

%;**sign

ificant

at5%

;***

sign

ificant

at1%

.



and public shareholders. The rationale for establishing them has been to create a reli-able, permanent delivery mechanism for MSME finance. Normally, 90 percent of themicrofinance banks’ loans have to be below a US$ 10,000 threshold, with the remain-ing 10 percent going up to about US$ 200,000. Because of the high fixed costs ofextending and monitoring loans, the larger loans account for a larger share of profits.However, the average size of the loans of these banks is kept well below the US$10,000 threshold.

The local banks, attracted by the profitability of the cash flow-based lendingfinanced through the EBRD loans, have gradually started utilizing their own fundsto provide cash flow loans. The assignment of clients of local banks to the cash flowprogramme is carried out at the branch level. As loan officers get trained in the cashflow methodology in a bank branch, all clients requesting very small loans are allo-cated to the cash flow programme. There is also no possibility of shifting specific cli-ents from one bank branch to another as MSMEs are allocated to branchesaccording to their geographical proximity to the branch. There is hence generally noincentive to allocate good quality clients to either the collateral or cash flowlending.13

The beneficiaries of cash flow programmes are private entrepreneurs and firms,ranging from self-employed one-person businesses to companies with up to 100employees. Loans start as low as US$ 20 (for example, for an open bakery on a Cen-tral Asian market to buy flour) up to about US$ 200,000 (for example, for the pur-chase of upholstery equipment for a furniture producer in Ukraine).

3. The survey, sample and basic statistics

During the first half of 2005, we administered a questionnaire to a sample of 1,272MSMEs (defined as firms with fewer than 250 employees) in Bulgaria, Georgia, Rus-sia and Ukraine.14 In each country, these MSMEs represent a stratified random

13 In one of the eight programmes, there was however a potential incentive for a bank to allocate loan appli-cants non-randomly between the cash flow and collateral loans. In the case of Hebros Bank in Bulgaria, theremay have been an incentive to shift the best clients from the collateral to the cash flow portfolio of loansbecause Hebros Bank received a credit line from EU/EBRD SME Facility to provide loans to MSMEs. EBRDoffered a discount on the interest charged on this credit line. We have therefore checked whether the estimatesbased on data from Hebros Bank differed from others and found that the Hebros and other estimates were notstatistically different from one another.14 The selection of countries in which the survey was run was based on a number of criteria: First, the numberof loans extended by each financial intermediary in the EBRD programme being at least 250; second, to allowfor a comparison of the impact on MSMEs of the different quality of finance provided by different types offinancial intermediaries, the presence of both dedicated microfinance institution and existing local banksadministering targeted credit lines; finally, in the case of large countries, an overlap of selected regions withregions in which the 2002 BEEPS was run, as the firms in the control group were selected from the BEEPS sam-ple. (BEEPS is a survey of over 9,500 companies in 26 transition countries and Turkey. See European Bank forReconstruction and Development (2005) for details.



sample of manufacturing, trade and other service sector enterprises that in 2002received finance from the EBRD’s MSME financial intermediaries (the treatmentgroup that represents roughly two thirds of the overall sample per country) andenterprises that by the time of the survey had not received finance from the EBRDintermediaries, but were in existence in 2002 (the control group comprising aboutone third of the overall sample per country). The treatment group firms wereselected as a random sample stratified by employment size and sector.15 The controlgroup comprises firms that are matched to the treatment group by location, employ-ment size and sector.

Although no quotas were applied to specific sectors, in practice most inter-viewed enterprises are in the (broadly defined) trade sector because a majority ofcompanies that borrowed from the banks in question are in this sector. Microenter-prises constitute the bulk of the loan portfolio. In the case of microfinance institu-tions and micro-lending programmes through participating banks, microenterprisesaccount on average for two thirds of the volume and 90 percent of the number ofloans.16 As the role of microenterprises in financial intermediaries’ portfolios is solarge, this is reflected in the size of sample strata by size class.17 Since we wanted toanalyse the impact of cash flow vs. collateral finance on enterprises of all sizes, weaimed at having all size classes represented in the sample. Yet, due to total samplesize limitations, in some cases the sample stratification does not match exactly thefinancial intermediaries’ portfolio composition, although it does reflect the domi-nance of micro firms in the banks’ portfolio by giving a larger weight to microenter-prises (54 percent of total number of surveyed enterprises) than to small (36 percent)or medium-sized enterprises (10 percent).

Table 2 shows the sample composition by size class and sector for both controland treatment groups. The control group firms were selected in 2005 as a stratifiedrandom sample from marketing lists, internet databases, yellow pages and inter-viewers’ walk-ins to match the treatment group in each country by categories oflocation, employment size and sector.18

15 Except for Ukraine it was not possible to find sufficient enterprises in the largest employment category asmost of the banks working for the EBRD did not extend a sufficient number of loans to this category of enter-prises. Also in the case of TUB in Georgia, it was impossible to interview the specified quota of 100 enterprisesper each bank due to the small number of loans extended by this bank in 2002 combined with business failuresand inability to reach the enterprises which benefited from TUB loans. This failure was compensated by add-ing more enterprises from the Procredit Bank in Georgia. In Bulgaria, the Hebros Bank and Procredit Bankhad several inaccurate contact entries and the sample was hence drawn with replacement.16 In the case of Hebros Bank, these data are unknown as monitoring of the use of the proceeds of the Facilityis based on its subloans’ size rather than on its subborrowers’ size.17 Quotas were specified for the size composition of the sample of enterprises to be interviewed (50 percent ofthe sample had to employ up to 9 employees, 20 percent between 10 and 24 employees, 15 percent between 25and 49 employees, and 15 percent between 50 and 249 employees).18 The matching was not on a one-to-one basis, but by categories of location, size and sector. All firms fromGeorgia and Russia came from one location – Tbilisi and Nizhniy Novgorod, respectively.



The summary statistics of the key variables used in our analysis are provided inTable 3. All variables have reasonable values and display considerable variationover time. Given that the matching of the control group to the treatment group wasstructured around location, employment size and sector of the firms, other variablesshow a larger variation. The summary statistics across size groups are presented inAppendix 1. The number of loans is evenly distributed across size classes for cashflow loans while more loans are granted to larger companies in the case of collateralloans. This is consistent with the notion that the traditional banking approach pre-fers to issue loans to bigger rather than smaller businesses. From Table 1, we can seethat collateral loans have a larger average loan size than cash flow loans in 2004, alsocollateral loans received by control firms in 2001 were larger than loans received bytreatment firms.

Table 2. Number of firms by employment and sector in 2004

Number of employees Sector

1–5 6–15 16–249 Trade Industry Service Total

BulgariaTreatment group 47 73 80 97 68 35 200Control group 25 42 53 34 41 45 120Total 72 115 133 131 109 80 320

GeorgiaTreatment group 115 58 31 139 34 31 204Control group 54 32 27 54 34 25 113Total 169 91 58 193 68 56 317

RussiaTreatment group 49 85 86 142 28 50 220Control group 38 42 35 75 25 15 115Total 87 127 121 217 53 65 335

UkraineTreatment group 68 56 76 103 80 17 200Control group 23 35 42 49 41 10 100Total 91 91 118 152 121 27 300

All countriesTreatment group 279 272 273 481 210 133 824Control group 140 151 157 212 141 95 448Total 419 423 430 693 351 228 1,272

Notes: Treatment group consists of MSMEs that had received a loan from one of eight EBRD-sponsoredMSME lending projects in 2002. Control group consists of MSMEs that had never received an EBRD projectloan.



Given our survey design, we do not observe any exit of firms. We do know thefailure rate among the firms that received EBRD credit in 2002. The failure rate from2002 to 2005 was less than 10 percent for firms in Bulgaria, Russia and Ukraine, andit was slightly above 10 percent for firms in Georgia. For our comparison group offirms, we do not have the failure statistics and we therefore cannot compare directlythe survival rates of the treatment and the control group of firms. However, to carryout this type of comparison, we selected another control group for this purpose,namely the enterprises that were surveyed by BEEPS survey in the 2002 and 2005waves. We can get the average attrition rates by country, which are around 15 per-cent, except for Georgia where the exit rate is 25 percent. In the BEEPS survey, wecan also split the firms into those that received credit and those that did not. Interest-ingly, we observe almost no failure among the credit-receiving Bulgarian and

Table 3. Summary statistics

Control group Treatment group

Mean Median SD Mean Median SD

2001Revenue 430 65 2,674 193 61 522Investment 14 0 64 12 2 48Fixed assets 117 9 759 55 9 207Net profit 44 7 146 54 10 286Total employment 18 7 35 12 5 26Leverage 7 0 16 7 0 15Number of firms with independent auditors 87 90Number of firms 401 779

2004Revenue 549 69 2,466 335 77 1,389Investment 26 1 125 24 3 89Fixed assets 125 11 538 87 14 308Net profit 64 8 234 114 13 1,160Total employment 19 8 33 19 8 34Leverage 8 0 17 14 4 21Number of firms with independent auditors 148 123Number of firms 448 824

Notes: Treatment group consists of MSMEs that had received a loan from one of eight EBRD-sponsoredMSME lending projects in 2002. Control group consists of MSMEs that had never received an EBRD projectloan. Financial data are expressed in thousands of US dollars. Figures are adjusted to producer prices. For allcountries the country specific producer price index is used except for Russia for which the Nizhny Novgorodregional producer price index is used. Total employment is a full time equivalent of full-time, part-time andtemporary employees. Leverage is defined as ratio of debt to debt plus equity.



Russian firms, while the failure rate among the credit receivers in Georgia and Uk-raine is higher than the average (around 35 percent). Using the BEEPS sample as acontrol group has a significant drawback, however, in that we are not able to excludethe possibility that some of the loan recipients within the BEEPS group were clientsof the EBRD funded programmes for financing MSMEs. Overall, it seems that treat-ment firms are less prone to failure. Hence, our sampled firms may be better per-formers than the average firm in a country (this distinction is even more pronouncedwith respect to the control group of the firms since their failure rates were higher).

4. Analytical framework

In carrying out our analysis, we need to take into account the fact that our sampledfirms differ as to whether they received a cash flow loan in 2002 and also whetherand when they received other loans. In particular, firms in the treatment group mayhave received other cash flow or collateral loans before and after 2002, while firmsin the control group may have received collateral loans at any time. From an analyti-cal standpoint there may hence be a selection problem, with better performing firmsfor instance being more able to obtain cash flow and/or collateral loans. If one didnot control for this non-random assignment of firms to loans, one could mistakenlyattribute the superior post-2002 performance to loans rather than recognizing thatpart may be due to inherently superior performance of the firms that receive loans.In view of the design of our sample, we are able to control for the treatment and per-formance of different firms up to 2002, and then focus on analysing the impact ofsubsequent cash flow and collateral loans on performance.

4.1 Determinants of receiving a loan

The probability that a firm receives a cash flow or collateral loan, respectively, ispredicted by two dummy variables indicating whether the firm received a cash flowor collateral loan 2 years earlier, a dummy variable reflecting whether the firm hadan independent auditor 2 years before a loan was granted, and a continuous vari-able reflecting the firm’s initial leverage in 2001:19

CFit ¼ aþ bCFit�2 þ cCLit�2 þ dAuditorit�2 þ /Leveragei2001 þ eit; ð1Þ

CLit ¼ aþ bCFit�2 þ cCLit�2 þ dAuditorit�2 þ /Leveragei2001 þ eit; ð2Þ

where CFit (CLit) is a dummy variable which assumes value 1 if a cash flow (collateral)loan is awarded to firm i in year t and 0 in all other cases, Auditor is a dummy variable

19 Using dummy variables for the initial (2000 or 2001) loan status instead of the 2-year lagged loan statusyields similar results.



equal to 1 if firms had an independent auditor, Leverage is a ratio of debt to debt plusequity and eit is an error term. We also control for country effects, industry effects andtime effects. We apply a logit estimation procedure to evaluate Equations (1) and (2).

In alternative specifications, we use the amount of loan received as a dependentvariable and we use the same set of determinants to predict the size of loan received.We apply a Tobit estimation procedure.

4.2 The effects of loans on performance

Formally, in the spirit of Ashenfelter and Card (1985), Heckman and Hotz (1989),and Hanousek et al. (2007), we specify a panel-data procedure. Let Xit be a givenperformance indicator, with subscript i denoting an individual firm and t denotingyear (t goes from 2001 to 2004). A simple model of performance may be specified inthe form of an annual rate of change (first-difference of logs) of the dependent vari-able as

D lnXit ¼ aþ b lnXi2001 þ cCFit�1 þ dCLit�1 þ eit; ð3Þ

where eit is the error term. Our interest is in estimating the c of cash flow loans andd of collateral loans obtained in the 2002–2004 period. In empirical estimations ofEquation (3), we also control for country effects, industry effects and time effects.Note that Equation (3) is relatively flexible and that it takes firms that received noloans as the base, permitting their percentage change in performance to vary overtime at the rate a + b lnXi2001. In addition, Equation (3) controls for the effects onperformance of any fixed differences among all firms.20 A particular concern is thatwe should ensure that our estimates capture the effect of loans rather than otherfactors such as competition. As may be seen from Equation (3), we do so by con-trolling for these other factors by the initial (2001) performance and by includingthe aforementioned country, industry and time effects. Finally, we also allow fortwo specifications of the effect of the two types of credit: one where the effect doesnot vary with the amount of credit and one where the effect of credit varies withloan size.

There are three key econometric issues that we need to account for in our analy-sis: omitted variables bias, measurement error, and endogeneity/selection of receiv-ing loans. We address omitted variables bias by including a number of importantcontrol variables described above. With respect to measurement error in loans, per-formance and other variables, we note that the error can induce attenuation bias aswell as more complicated biases in estimated coefficients. Being aware of this prob-lem, in collecting the dataset we placed particular emphasis on identifying precisely

20 Note that we have also estimated the effects of loans on the level (as opposed to the rate of change) of per-formance and found these effects to be statistically insignificant and their exclusion not to materially affect theother parameter estimates.



individual loans, as well as carefully collecting several indicators of performance forthe current and preceding periods. In this respect, our survey is of higher qualitythan many other surveys in this area. Our emphasis on collecting high quality infor-mation is also reflected in the relatively high response rate (38 percent) that we havegenerated from firms for our questionnaires. Finally, we also checked that there areno outliers that would seriously affect our estimates.

As to endogeneity/selection of receiving loans, there is a danger that the inher-ently superior performance of the firms selected for receiving cash flow or collateralloans could be attributed to loans rather than the possibly non-random assignmentof firms to loans. We address this problem as follows. First, we match the controlgroup firms with the treatment group on the three observable characteristics dis-cussed above. Second, we use the panel data specification in Equation (3) with theaforementioned covariates as controls. This controls for the possibility that firms arenot assigned to loans at random and that lending institutions may give loans tofirms that are inherently superior or inferior performers. In addition, our surveyincludes questions with regard to business constraints. One of the constraints thatfirms were asked about was the availability of financing from banks. The responseto this question was statistically indifferent for treatment and control group firms.This allows us to assume that fewer loans in the control group were the firms’choice. Also, the treatment firms considered the cost of financing as a more severeconstraint than did the control firms. This again suggests that control firms were nottreated worse than the treated firms with respect to firm financing.

5. The empirical results

We present our empirical estimates in three parts. First, we discuss the resultsrelated to the determinants of the probability that a firm receives a cash flow or col-lateral loan. Second, we examine the effects of the presence of cash flow and collat-eral loans on MSME performance, irrespective of the size of these loans. We carryout this estimation for all firms together and separately by firm size, using the threesize categories of firms corresponding to micro, small and medium-sized firms.21

Finally, we examine the extent to which the effects of loans vary by loan size.In Panel A of Table 4, we report marginal effects of a logit estimation relating the

probability of receiving a cash flow or collateral loan in a given year to the explana-tory variables in Equations (1) and (2). In Panel B, we report Tobit estimates relatedto the size of the loan as the dependent variable. In all tables, we first present

21 The firms are split into size classes based on employment in year 2002. Our data show that allowing all theregression coefficients to be different (i.e. running separate regressions) in the three size categories of firms isstatistically superior to constraining all coefficients to be the same and allowing only the loan effect to vary byfirm size (i.e. by interacting loan with size). We therefore run separate regressions. The two approaches yieldsimilar results.



Tab

le4.

Pan

elA

–determinan

tsof

receivingaloan

(marginal

effectsfrom

logite

stim

ations);P

anel

B–determi-

nan

tsof

aloan

size

received

(Tob

itestimations)

Pan

elA

Allfirm

sM

icro

firm

sSmallfi

rms

Med

ium

firm

s

Cashflow

loan

Collateral

loan

Cashflow

loan

Collateral

loan

Cashflow

loan

Collateral

loan

Cashflow

loan

Collateral

loan

Cashflow

loan

(t�2

)0.48

6(0.020

)***

�0.031

(0.007

)***

0.50

7(0.034

)***

�0.031

(0.008

)***

0.45

8(0.035

)***

�0.019

(0.012

)*0.47

4(0.041

)***

�0.062

(0.021

)***

Collateralloa

n(t�2

)�0

.114

(0.040

)***

0.32

5(0.037

)***

�0.051

(0.111

)0.15

5(0.068

)**

0.01

7(0.074

)0.26

5(0.069

)***

�0.208

(0.051

)***

0.45

4(0.055

)***

Anindep

enden

tau

dito

r(t�2

)�0

.146

(0.025

)***

0.03

3(0.012

)***

�0.004

(0.058

)0.01

1(0.017

)� 0

.147

(0.048

)***

0.03

8(0.023

)�0

.202

(0.039

)***

0.01

1(0.023

)Lev

erag

ein

2001

0.05

0(0.060

)0.05

9(0.018

)***

0.09

3(0.122

)0.05

6(0.022

)**

0.15

0(0.096

)0.00

4(0.035

)�0

.062

(0.107

)0.13

2(0.046

)***

Observa

tions

3,78

33,78

31,41

31,32

91,31

71,29

31,05

31,01

7Ps

eudoR2

0.16

0.21

0.15

0.12

0.16

0.14

0.21

0.28

F-statistic

423.26

309.14

134.81

44.08

139.04

56.84

138.39

152.01



Pan

elB

Allfirm

sM

icro

firm

sSmallfi

rms

Med

ium

firm

s

Cashflow

loan

Collateral

loan

Cashflow

loan

Collateral

loan

Cashflow

loan

Collateral

loan

Cashflow

loan

Collateral

loan

Cashflow

loan

(t�2

)8.32

6(0.386

)***

�5.144

(1.278

)***

8.00

6(0.616

)***

�9.583

(3.223

)***

7.48

5(0.623

)***

�3.709

(2.299

)8.52

9(0.790

)***

�4.825

(1.821

)***

Collateralloa

n(t�2

)�2

.102

(0.713

)***

18.551

(1.554

)***

�1.065

(1.482

)16

.242

(4.557

)***

0.53

6(1.172

)18

.392

(3.457

)***

�4.838

(1.228

)***

16.249

(1.732

)***

Anindep

enden

tau

dito

r(t�2

)�2

.188

(0.452

)***

4.52

1(1.234

)***

�0.103

(0.905

)1.89

1(3.788

)� 2

.049

(0.777

)***

6.36

6(2.771

)**

�4.002

(0.828

)***

1.15

3(1.476

)Lev

erag

ein

2001

0.85

2(0.951

)7.65

6(2.685

)***

1.35

5(1.773

)13

.247

(7.365

)*2.16

9(1.459

)�0

.175

(5.559

)�1

.146

(1.840

)7.32

4(3.176

)**

Observa

tions

3,78

33,78

31,41

31,41

31,31

71,31

71,05

31,05

3Ps

eudoR2

0.05

0.10

0.05

0.08

0.05

0.08

0.07

0.13

F-statistic

136.46

50.34

45.75

6.56

41.36

9.25

44.43

29.77

Notes:D

epen

den

tvariables

arecash

flow

(collateral)loan

dum

mywhich

equa

ls1iffirm

received

cash

flow

(collateral)loan

inagive

nye

ar(Pan

elA),an

dcash

flow

(collateral)loan

dum

myinteracted

with

thelogloan

size

(inUSdollars)(Pa

nelB

).Anindep

enden

taud

itorisadum

myva

riab

le,

which

equa

ls1iffirm

hadtheinde

pend

entau

dito

rat

least2ye

arsbe

fore

receivingaloan

.Lev

erag

eis

aratio

ofdeb

tov

erthesu

mof

deb

tan

deq

uity.A

llregression

sinclud

eindus

try,

coun

tryan

dye

ardum

mies.Micro

firm

sha

ve1–5,

smallfi

rmsha

ve6–15

andmed

ium-sized

firm

sha

ve16

ormoreem

ploy

ees.Rob

uststan

darderrors

(Pan

elA)an

dstan

darderrors

(Pan

elB)aregive

nin

parenthe

ses.*significant

at10

%;**significant

at5%

;***sign

ificant

at1%

.



estimates for all firms taken together and then separate estimates for each of thethree firm size categories.22

As may be seen from Table 4, the 2-year lagged loan variables have a strongexplanatory power, indicating that firms that received a given (cash flow or collat-eral) loan in the past are more likely to receive the same type of loan, and also a lar-ger size loan of the same type, in the future. The effect of having received a cashflow loan in the past has a negative effect on receiving a collateral loan in the future,however, and this effect is significant for all firm size categories. The same negativecross-effect is seen for past collateral loans on current cash flow loans for the sampleas a whole and for medium-sized firms, but it is insignificant for micro and smallfirms. We discuss the implication of these findings below.

The presence of an independent auditor increases the probability that firmsreceive collateral loans and it also has a positive effect on the size of collateral loans,although this positive effect is statistically insignificant in some of the subsamples.On the other hand, the presence of an independent auditor reduces the probabilitythat firms receive a cash flow loan and the size of this type of a loan, with this nega-tive effect also not being statistically significant in some of the subsamples. Overall,it appears that banks that rely on the collateral method of assessing creditworthinessview favourably the presence of independent auditors, while banks relying on thecash flow method tend to ignore or even discount the presence of independent audi-tors.23 Leverage in 2001 has a positive effect on the probability of firms getting a col-lateral loan and on collateral loan size. On the whole, the explanatory variables havea strong explanatory power in both sets of regressions. The pseudo R2s are in the0.12–0.28 and 0.07–0.13 range, respectively.

The effects of loans on performance are presented in Tables 5–8. In each table,we first give estimates from a model in which the performance effects of loans donot vary with loan size and subsequently estimates from a model where the effect ofcredit varies with loan size.

In Table 5, we report the effects of cash flow and collateral loans on the rate ofgrowth of fixed assets. In the first column of the table, the estimated average effectsof a loan, based on data for all firms, suggest that the award of a cash flow loan (col-lateral loan) is related to a 10.7 (15.7) percentage point higher growth rate of fixedassets of the firm. Since the collateral loans are on average three times as large as thecash flow loans, the percentage effect per dollar of loan may be argued to be abouttwice as high for the cash flow loans than collateral loans.24 However, since firmsreceiving collateral loans are on average about twice as large as firms receiving cash

22 Note that these size categories give us a similar number of observations for each group and hence providea useful stratification for drawing inferences about the probability of obtaining a loan and the effects of loansin micro, small and medium-sized enterprises.23 Alternatively, it could be that firms with an independent auditor tend to be firms with good collateral andhence have lesser need for cash flow loans. We thank the Editor for pointing this out.24 The intuition here is that the 15.7 percentage effect would become one third (5.23 percentage effect) if thesize of the collateral loan were just one third of its size.



Tab

le5.

Loa

neffectson

fixedassets

Firm

s

All

Micro

Small

Med

ium

All

Micro

Small

Med

ium

Cashflow

loan

0.107

(0.019)***

0.052

(0.026)**

0.106

(0.026)***

0.174

(0.042)***

�0.736

(0.161)***

�0.481

(0.143)***

�0.507

(0.167)***

�1.238

(0.489)**

Collateralloa

n0.157

(0.032)***

0.203

(0.091)**

0.114

(0.046)**

0.163

(0.047)***

�0.383

(0.165)**

�0.586

(0.466)

�0.656

(0.351)*

�0.511

(0.327)

Log

fixedassetsin

2001

�0.036

(0.006)***

�0.032

(0.011)***

�0.060

(0.010)***

�0.065

(0.021)***

�0.053

(0.008)***

�0.042

(0.013)***

�0.068

(0.010)***

�0.084

(0.027)***

Cashflow

loan

9logloan

size

0.095

(0.019)***

0.066

(0.019)***

0.069

(0.018)***

0.141

(0.052)***

Collateralloa

n9

logloan

size

0.057

(0.017)***

0.094

(0.062)

0.082

(0.037)**

0.065

(0.032)**

Critic

allogcash

flow

loan

size

7.752

(0.201)***

7.246

(0.368)***

7.387

(0.586)***

8.789

(0.330)***

Corresp

ondingpe

rcen

tile

from

theloan

size

distribution

2125

921

Critic

allogcolla

teralloa

nsize

6.689

(0.992)**

6.22

(1.125)***

8.023

(0.806)***

7.807

(1.365)**

Corresp

ondingpe

rcen

tile

from

theloan

size

distribution

312

65

Observa

tions

3,605

1,305

1,270

1,030

3,605

1,305

1,270

1,030

R2

0.08

0.08

0.17

0.08

0.11

0.09

0.19

0.11

Notes:D

epen

den

tvariableisdefi

nedln(X

t)�l

n(Xt�

1).C

ashflow

(collateral)loan

dum

myeq

uals1iffirm

received

cred

itba

sedon

cash

flow

(col-

lateral)metho

din

agive

nye

ar.T

hecriticallog

cash

flow

(collateral)loan

size

iscalculated

bydividingthecoefficien

tofc

ashflow

(collateral)loan

bythecoefficien

tofc

ashflow

(collateral)loan

9logloan

size.A

llregression

sinclud

eindus

try,

coun

tryan

dye

ardum

mies.Micro

firm

sha

ve1–5,

smallfi

rmsha

ve6–

15,a

ndmed

ium-sized

firm

sha

ve16

ormoreem

ploy

ees.Rob

uststan

darderrors

inpa

renthe

ses.*significant

at10

%;**signifi-

cant

at5%

;***sign

ificant

at1%

.



Tab

le6.

Loa

neffectson

reve

nues

Firm

s

All

Micro

Small

Med

ium

All

Micro

Small

Med

ium

Cashflow

loan

0.045

(0.012)***

0.047

(0.018)**

0.040

(0.019)**

0.053

(0.026)**

�0.333

(0.061)***

�0.358

(0.114)***

�0.133

(0.123)

�0.442

(0.128)***

Collateralloa

n0.083

(0.022)***

0.175

(0.059)***

0.067

(0.030)**

0.030

(0.035)

�0.147

(0.135)

0.249

(0.272)

�0.598

(0.228)***

�0.415

(0.206)**

Log

reve

nuein

2001

�0.027

(0.004)***

�0.043

(0.008)***

�0.054

(0.009)***

�0.015

(0.010)

�0.035

(0.004)***

�0.048

(0.008)***

�0.058

(0.008)***

�0.025

(0.010)**

Cashflow

loan

9logloan

size

0.043

(0.007)***

0.05

(0.014)***

0.019

(0.014)

0.049

(0.013)***

Collateralloa

n9

logloan

size

0.024

(0.014)*

�0.009

(0.033)

0.070

(0.024)***

0.043

(0.020)**

Critic

allogcash

flow

loan

size

7.809

(0.303)***

7.113

(0.425)***

6.874

(1.689)***

8.956

(0.545)***

Corresp

ondingpe

rcen

tile

from

theloan

size

distribution

2319

325

Critic

allogcolla

teralloa

nsize

6.016

(2.214)***

28.533

(78.365)

8.518

(0.525)***

9.681

(0.788)***

Corresp

ondingpe

rcen

tile

from

theloan

size

distribution

299

1927

Observa

tions

3,686

1,343

1,297

1,046

3,686

1,343

1,297

1,046

R2

0.10

0.13

0.15

0.08

0.11

0.14

0.16

0.09

Notes:D

epen

den

tvariableisdefi

nedln(X

t)�l

n(Xt�

1).C

ashflow

(collateral)loan

dum

myeq

uals1iffirm

received

cred

itba

sedon

cash

flow

(col-

lateral)metho

din

agive

nye

ar.T

hecriticallog

cash

flow

(collateral)loan

size

iscalculated


tofc

ashflow

(collateral)loan

bythecoefficien

tofc

ashflow

(collateral)loan

9logloan

size.A

llregression

sinclud

eindus

try,

coun

tryan

dye

ardum

mies.Micro

firm

sha

ve1–5,

smallfi

rmsha

ve6–15

,and

med

ium-sized

firm

sha

ve16

ormoreem

ploy

ees.Rob

uststan

darderrors

inpa

renthe

ses.*significant

at10

%;**signifi-

cant

at5%

;***sign

ificant

at1%

.



Tab

le7.

Loa

neffectson

employm

ent

Firm

s

All

Micro

Small

Med

ium

All

Micro

Small

Med

ium

Cashflow

loan

0.07

7(0.013

)***

0.05

3(0.022

)**

0.05

6(0.019

)***

0.07

8(0.023

)***

�0.426

(0.064

)***

�0.223

(0.108

)**

�0.325

(0.111

)***

�0.282

(0.164

)*Collateralloa

n0.14

(0.026

)***

0.18

4(0.080

)**

0.15

1(0.044

)***

0.07

4(0.029

)**

�0.179

(0.118

)�0

.633

(0.476

)0.08

2(0.266

)�0

.558

(0.157

)***

Log

employ

men

tin

2001

�0.073

(0.008

)***

�0.141

(0.021

)***

�0.21

(0.030

)***

�0.166

(0.035

)***

�0.089

(0.009

)***

�0.145

(0.021

)***

�0.212

(0.030

)***

�0.176

(0.035

)***

Cashflow

loan

9logloan

size

0.05

6(0.007

)***

0.03

4(0.014

)**

0.04

2(0.012

)***

0.03

6(0.016

)**

Collateralloa

n9

logloan

size

0.03

4(0.012

)***

0.09

7(0.062

)0.00

7(0.027

)0.06

(0.015

)***

Critic

allogcash

flow

loan

size

7.57

3(0.269

)***

6.54

3(0.753

)***

7.68

5(0.569

)***

7.90

4(1.206

)***

Corresp

ondingpe

rcen

tilefrom

theloan

size

distribution

198

155

Critic

allogcolla

teralloa

nsize

5.30

7(1.734

)***

6.49

4(0.975

)***

9.23

3(0.517

)***

Corresp

ondingpe

rcen

tilefrom

theloan

size

distribution

113

15

Observa

tions

3,72

41,34

61,31

51,05

13,72

41,35

81,31

51,05

1R2

0.08

0.09

0.21

0.21

0.10

0.10

0.21

0.21

Notes:D

epen

den

tva

riab

leisdefi

nedln(X

t)�l

n(Xt�

1).C

ashflow

(collateral)loan

dum

myeq

uals1iffirm

received

cred

itba

sedon

cash

flow

(col-

lateral)metho

din

agive

nye

ar.T

hecriticallog

cash

flow

(collateral)loan

size

iscalculated


tofc

ashflow

(collateral)loan

bythecoefficien

tofc

ashflow

(collateral)loan

9logloan

size.A

llregression

sinclud

eindus

try,

coun

tryan

dye

ardum

mies.Micro

firm

sha

ve1–

5,sm

allfi

rmsha

ve6–15

,and

med

ium-sized

firm

sha

ve16

ormoreem

ploy

ees.Rob

uststan

darderrors

inpa

renthe

ses.*significant

at10

%;**signifi-

cant

at5%

;***sign

ificant

at1%

.



Tab

le8.

Loa

neffectson

net

profit

Firm

s

All

Micro

Small

Med

ium

All

Micro

Small

Med

ium

Cashflow

-based

loan

0.09

6(0.025

)***

0.11

6(0.038

)***

0.09

8(0.042

)**

0.06

1(0.057

)�0

.226

(0.113

)**

�0.251

(0.199

)�0

.041

(0.252

)�0

.230

(0.260

)Collateral-b

ased

loan

0.09

2(0.045

)**

0.14

3(0.101

)0.18

5(0.078

)**

�0.050

(0.073

)�0

.302

(0.216

)�0

.363

(0.525

)�0

.821

(0.410

)**

�0.582

(0.323

)*Log

netp

rofitin20

01�0

.048

(0.008

)***

�0.068

(0.016

)***

�0.082

(0.015

)***

�0.035

(0.016

)**

�0.055

(0.009

)***

�0.075

(0.017

)***

�0.085

(0.015

)***

�0.042

(0.017

)**

Log

loan

size

*Cash

flow

-based

loan

0.03

6(0.013

)***

0.04

5(0.025

)*0.01

6(0.028

)0.02

9(0.025

)Log

loan

size

*Collateral-

basedloan

0.04

1(0.022

)*0.06

0(0.062

)0.10

6(0.043

)**

0.05

1(0.030

)*Critic

allogcash

flow

-ba

sedloan

size

6.24

7(1.099

)***

5.52

2(1.525

)***

2.60

8(11.55

5)7.93

2(2.713

)***

Corresp

ondingpe

rcen

tilefrom

theloan

size

distribution

21

985

Critic

allogcolla

teral-b

ased

loan

size

7.37

4(1.640

)***

6.00

7(2.861

)**

7.76

6(0.951

)***

11.475

(1.572

)***

Corresp

ondingpe

rcen

tilefrom

theloan

size

distribution

89

578

Observa

tions

3,44

41,25

31,21

497

73,44

41,25

31,21

497

7R2

0.04

0.05

0.07

0.03

0.04

0.05

0.07

0.04

Notes:D

epen

den

tvariableisdefi

nedln(X

t)�l

n(Xt�

1).C

ashflow

(collateral)loan

dum

myeq

uals1iffirm

received

cred

itba

sedon

cash

flow

(col-

lateral)metho

din

agive

nye

ar.T

hecriticallog

cash

flow

(collateral)loan

size

iscalculated


tofc

ashflow

(collateral)loan

bythecoefficien

tofc

ashflow

(collateral)loan

9logloan

size.A

llregression

sinclud

eindus

try,

coun

tryan

dye

ardum

mies.Micro

firm

sha

ve1–

5,sm

allfi

rmsha

ve6–

15,a

ndmed

ium-sized

firm

sha

ve16

ormoreem

ploy

ees.Rob

uststan

darderrors

inpa

renthe

ses.*significant

at10

%;**signifi-

cant

at5%

;***sign

ificant

at1%

.



flow loans, the effects of the two types of loans in terms of dollars of fixed assetsgenerated by a dollar of loan may be argued to be similar.25

The results based on firm size, reported in columns 2–4, indicate that both cashflow and collateral loans have a positive relationship with capital formation of allfirm sizes. Moreover, for cash flow loans the percentage effect rises and becomesmore statistically significant with firm size, while the relationship with collateralloans is more uniform. Overall, the results in the first part of Table 5 provide sup-port for the anecdotal assertion that the award of a (relatively short term) loan isassociated with higher fixed assets growth.

The estimated effects of loan size on fixed assets are reported in the last four col-umns of Table 5. The effects of cash flow loans on fixed assets vary systematicallywith the size of the loan in all three types of firms, with the effect being negative forsmall loans and becoming positive as loan size increases. The same pattern isobserved for collateral loans in small and medium-sized firms, while the effect inmicro firms does not vary significantly with loan size. As the calculated critical(overtaking) values in the table indicate, for the sample as a whole the effect of cashflow loans (collateral loans) on fixed assets turns from negative to positive at the21st (3rd) percentile of the cash flow loan (collateral loan) size distribution.26 Theestimates from regressions based on firm size in turn suggest that the critical cashflow loan values for micro, small and medium-sized firms are at the 25th, 9th and21st percentile, respectively. The corresponding critical values of collateral loans forsmall and medium-sized firms are at the 6th and 5th percentile, respectively. Theresults hence suggest that most cash flow and collateral loans yield a positive rela-tionship with growth of fixed assets. In most cases, the estimates also raise the issuethat very small loans may have a negative association with fixed asset formation. Inall regressions, the effect of the initial level of fixed assets is negative, as expected,indicating that the data display conversion to the mean – a phenomenon that weobserve with respect to the other performance variables in the following tables aswell.

The average effects of cash flow and collateral loans on firm revenues andemployment growth are reported in Tables 6 and 7. On average, receiving a cashflow loan is related to 4.5 percent higher rate of growth of revenue than would bethe case if the firm did not receive such a loan. The average effect of a collateral loanis estimated at 8.3 percent. Both cash flow and collateral loans have a positive rela-tionship with the rate of growth of employment as well (7.7 and 14 percent, respec-tively). The estimated effects of cash flow and collateral loans in regressions basedon firm size (columns 2–4) are positive and mostly statistically significant for bothperformance measures. The estimated effects of loans on revenues and employment

25 The effect may be argued to be similar in terms of dollars of fixed assets in the sense that the effect (of anidentically sized loan) is 5.23 percentage points for a firm with fixed assets twice as sizable as the fixed assetscorresponding to the firm facing the 10.7 percentage point loan effect.26 The critical loan size values are reported only if either or both the loan dummy and the interaction of loandummy and loan size variable are statistically significant.



are hence broadly consistent with the corresponding positive effect of loans on fixedassets and they suggest that loan receivers become larger.

The estimated effects of the loan size on the rate of change of revenues andemployment indicate that the effects of loans are negative for small loans andbecome positive as loan size increases. As with the effect on fixed assets, the resultsacross size categories indicate that most cash flow and collateral loans yield a posi-tive effect on the growth of revenues and employment, but that in some firm sizecategories the effect of the smallest loans may be negative. In estimations that we donot report here, we have also found a positive effect of loans on labour costs. Theestimated coefficients from the labour cost regressions imply that the overall labourcost effect of loans is primarily accounted for by the positive effect of loans onemployment rather than on wages (we can derive wage per employee from labourcost and we can show that this was not affected by the loans received).27

Since revenues and costs are the two principal components of profit, we havealso examined directly the effect of loans on profit (see Table 8). We find that theeffect of the two types of loans on profitability is also positive and statistically signif-icant (1st column of Table 8). Our results with respect to revenues and profit hencesuggest that receiving a loan is related to increased scale of operations and higherprofits.

6. Conclusions

Our analysis of micro, small and medium-sized enterprises (MSMEs) permits usto draw two sets of interesting conclusions, one with respect to the allocation ofcredit and one with respect to the effects on firm performance of two types ofloans – a new type of loan based on cash flow that was pioneered and spear-headed by EBRD in the transition economies and a traditional-style loan basedon collateral.

With respect to the allocation of credit, we find that MSMEs that received acash flow or collateral loan in the past are more likely to receive the same type ofloan, and also a larger sized loan of the same type, in the future. This finding mayreflect the fact that firms that received a given loan in the past are more likely toapply to the same lender and/or that the lender is more likely to lend to firms towhich he/she lent in the past. Interestingly, the effect of having received a cashflow loan in the past has a negative effect on the probability of receiving a collat-eral loan in the future, while the same negative cross-effect from receiving collat-eral loans in the past on the probability of receiving a cash flow loan is statistically

27 A separate set of results is received by analysing the capital–labour ratio, which was proxied by the fixedassets to total employment ratio. We found that receiving a loan has no significant effect on the change of thisvariable. This confirms the finding that both fixed assets and employment are affected positively by loans andwe cannot find a larger tendency in one or in the other. These results are available on request.



insignificant for micro and small firms. These results suggest that we observe amixture of two effects. First, it appears that cash flow loans are the preferred formof credit in the sense that once MSMEs of any size receive this type of loan theyare more likely to receive another in the future and less likely to obtain a collateralloan in the future. This hypothesis is also supported by the finding that micro andsmall firms are unaffected in their probability of obtaining a cash flow loan in thefuture by having received a collateral loan in the past. The second effect seems tobe that banks and clients develop a specific relationship that makes them morelikely to deal with each other over time and less likely to switch to another partner(relationship banking). This may be viewed as a form of market segmentation andit is consistent with findings based on US small and medium-sized firms. In partic-ular, Petersen and Rajan’s (1994) findings suggest that bank relationships are valu-able to small firms.

With respect to the effect of credit on performance, we show that both cashflow loans and collateral loans are positively related to a number of key perfor-mance indicators of small and medium-sized enterprises. In the case of microenter-prises, the association of these loans with performance is also found to begenerally positive but somewhat less significant. Both types of loans have an over-all positive relationship with fixed asset formation, suggesting that firms use bothtypes of bank loans for investment in fixed capital. In terms of dollars of fixedassets generated by a dollar of loan, the effects of the two types of loans aresimilar.

Both cash flow and collateral loans also by and large display a positive relation-ship with revenues and employment. In particular, the overall estimates for all firmsindicate that the loans enable the MSMEs to expand production beyond the scalethat they could have achieved without this source of credit. Finally, there is also apositive relationship between both types of loan and profitability.

Our estimates also suggest that while cash flow and collateral loans are posi-tively related to various performance indicators, in many cases the performanceeffects of the smallest size loans may be negative. This could be interpreted in twodifferent ways. First, that the microenterprises awarded with the smallest size loansare the poorest and use loan proceeds for purposes defined as wasteful from anenterprise point of view – that is, loan proceeds may be used to finance consump-tion/survival of the entrepreneur’s household. Second, that the microenterprisesawarded with the smallest loans are first-time borrowers who are not as experiencedor as successful entrepreneurs as those who are awarded larger loans, and thusmight be less efficient in the use of funds made available to them. This finding hasimplications for lending policies and deserves more in-depth investigation in futureresearch.

Finally, from the policy standpoint it is important to note that it is easier and fas-ter for MSMEs to qualify for the cash flow loans than collateral loans. In our sample,we also see that most of the collateral loans obtained by control firms have beenreceived by medium-sized firms, while the allocation of cash flow loans was



relatively even across all the firm size groups. This suggests that cash flow loans canplay an important part in credit provision to micro and small enterprises that maybe overlooked by the traditional collateral-based loan providers. Overall, our dataand analysis suggest that the EBRD spearheaded programme of cash flow loans hasbeen a success.

References

Alexander-Tedeschi, G. and Karlan, D. (2009). ‘Cross sectional impact analysis: Bias fromdropouts’, Perspectives of Global Development and Technology, 9, pp. 270–291.

Ashenfelter, O. and Card, D. (1985). ‘Using the longitudinal structure of earnings to estimatethe effect of training programs’, Review of Economics and Statistics, 67, pp. 648–660.

Bah, E., Brada, J. C. and Yigit, T. (2011). ‘With the help from our friends: The effect of USAIDassistance on SME growth in a transition economy’, Journal of Comparative Economics, 39,pp. 205–220.

Banerjee, A. V. and Duflo, E. (2004). ‘Do firms want to borrow more? Testing credit con-straints using a direct lending program’, mimeo, MIT.

Berger, A. N. and Udell, G. F. (2006). ‘A more complete conceptual framework for financingof small and medium enterprises’, Journal of Banking and Finance, 30, pp. 2,945–2,966.

Brown, J. D., Earle, J. and Lup, D. (2002). ‘Determinants of small firm growth: Finance, humancapital, technical assistance and the business environment’, Heriot-Watt University andCentral European University, mimeo.

Coleman, B. (1999). ‘The impact of group lending in northeast Thailand’, Journal of Develop-ment Economics, 60, pp. 105–141.

European Bank for Reconstruction and Development (2005). Transition Report 2005, Business inTransition, London: EBRD.

Gin�e, X., Harigaya, T., Karlan, D. and Nguyen, B. T. (2006). ‘Evaluating microfinance programinnovation with randomised control trials: An example from group versus individuallending’, ERD Technical Note Series, N. 16, March 2006, Asian Development Bank,Manila.

Hanousek, J., Ko�cenda, E. and Svejnar, J. (2007). ‘Origin and concentration. Corporateownership, control and performance in firms after privatization’, Economics of Transition,15, pp.1–31.

Heckman, J. J. and Hotz, V. J. (1989). ‘Choosing among alternative nonexperimental methodsfor estimating the impact of social programs: The case of manpower training’, Journal ofthe American Statistical Association, 84, pp. 862–874.

Hulme, D. and Mosley, P. (1996). Finance Against Poverty, 2 vols. London: Routledge.Karlan, D. S. (2001). ‘Microfinance impact assessments: The perils of using new members as a

control group’, Journal of Microfinance, 3, pp. 75–85.Karlan, D. S. and Goldberg, N. (2006). ‘The impact of microfinance: A review of methodologi-

cal issues’, World Bank, Doing Impact Evaluation Series, 7, Washington, DC: The WorkdBank.

Krasniqi, B. A. (2010). ‘Are small firms really credit constrained? Empirical evidence fromKosovo’, International Entrepreneurship and Management Journal, 6, pp. 459–479.



LeLarge, C., Sraer, D. and Thesmar, D. (2008), ‘Entrepreneurship and credit constraints: Evi-dence from a French loan guarantee program’, in Lerner J., Schoar A. (eds.) InternationalDifferences in Entrepreneurship, Chicago: University of Chicago Press, pp. 243–273.

Morduch, J. (1999). ‘The microfinance promise’, Journal of Economic Literature, 37, pp. 1,569–1,614.

Mosley, P. (1998). ‘The use of control groups in impact assessments for microfinance’, Work-ing paper no. 19, International Labour Office: Geneva

Petersen, M. A. and Rajan, R. G. (1994). ‘The benefits of lending relationships: Evidence fromsmall business data’, Journal of Finance, 49, pp. 3–37.

Pissarides, F., Singer, M. and Svejnar, J. (2003). ‘Objectives and constraints of entrepreneurs:Evidence from small and medium size enterprises in Russia and Bulgaria’, Journal of Com-parative Economics, 31, pp. 503–531.

Storey, D. J. (2000). ‘Six steps to heaven: Evaluating the impact of public policies to supportsmall businesses in developed economies’, in Landstrom H., Sexton D. L. (eds.), Handbookof Entrepreneurship, Oxford: Blackwells, pp. 176–194.



Appendix

1

Pan

elA

–Summarystatistics

formicro

firm

s(1–5

employe

es)

Con

trol

grou

pTreatmen

tgroup

Mea

nM

edian

SD

Mea

nM

edian

SD

2001 Rev

enue

4216

7499

2745

5Inve

stmen

t2

05

51

12Fixedassets

72

2334

321

8Net

profi

t8

317

406

332

Total

employ

men

t3

31

33

1Lev

erag

e3

012

50

12Num

berof

firm

swith

indep

enden

taud

itors

1219

Num

berof

cash

flow

loan

s91

Cashflow

loan

size

52

7Num

berof

colla

teralloa

ns6

9Collateralloa

nsize

118

144

33

Num

berof

firm

s14

633

020

04 Rev

enue

4116

7851

3168

Inve

stmen

t2

05

51

15Fixedassets

83

1414

329

Net

profi

t6

312

127

19Total

employ

men

t3

31

33

1Lev

erag

e4

011

111

18Num

berof

firm

swith

indep

enden

taud

itors

1923

Num

berof

cash

flow

loan

s15

9



*.(C

ontinued

)

Pan

elA

–Summarystatistics

formicro

firm

s(1–5

employe

es)

Con

trol

grou

pTreatmen

tgroup

Mea

nM

edian

SD

Mea

nM

edian

SD

Cashflow

loan

size

73

13Num

berof

colla

teralloa

ns10

8Collateralloa

nsize

114

1724

1530

Num

berof

firm

s14

027

9

Pan

elB–Summarystatistics

forsm

allfi

rms(6–1

5em

ploye

es)

Con

trol

grou

pTreatmen

tgroup

Mea

nM

edian

SD

Mea

nM

edian

SD

2001 Rev

enue

141

6627

617

177

378

Inve

stmen

t8

026

72

17Fixedassets

239

5047

1317

5Net

profi

t32

711

742

1219

1Total

employ

men

t9

93

109

3Lev

erag

e7

017

100

17Num

berof

firm

swith

indep

enden

taud

itors

2231

Num

berof

cash

flow

loan

s94

Cashflow

loan

size

95

11Num

berof

colla

teralloa

ns8

19Collateralloa

nsize

3917

4711

512

Num

berof

firm

s12

427

020

04 Rev

enue

166

6238

115

477

293

Appen

dix

1.(C

ontinued

)



*.(C

ontinued

)

Pan

elB–Summarystatistics

forsm

allfi

rms(6–15em

ploye

es)

Con

trol

grou

pTreatmen

tgroup

Mea

nM

edian

SD

Mea

nM

edian

SD

Inve

stmen

t5

011

93

20Fixedassets

288

6450

1717

3Net

profi

t31

1085

3013

91Total

employ

men

t10

93

1010

3Lev

erag

e9

017

145

19Num

berof

firm

swith

indep

enden

taud

itors

3144

Num

berof

cash

flow

loan

s15

6Cashflow

loan

size

167

25Num

berof

colla

teralloa

ns28

15Collateralloa

nsize

159

1620

827

Num

berof

firm

s15

127

2

Pan

elC–Summarystatistics

formed

ium-sized

firm

s(16or

moreem

ploye

es)

Con

trol

grou

pTreatmen

tgroup

Mea

nM

edian

SD

Mea

nM

edian

SD

2001 Rev

enue

1,13

621

84,60

239

915

272

9Inve

stmen

t35

510

531

794

Fixedassets

328

501,30

510

732

223

Net

profi

t94

2822

096

2731

0Total

employ

men

t47

2651

4126

45Lev

erag

e18

220

90

17Num

berof

firm

swith

indep

enden

taud

itors

5637

Appen

dix

1.(C

ontinued

)



*.(C

ontinued

)

Pan

elC–Summarystatistics

formed

ium-sized

firm

s(16or

moreem

ploye

es)

Con

trol

grou

pTreatmen

tgroup

Mea

nM

edian

SD

Mea

nM

edian

SD

Num

berof

cash

flow

loan

s67

Cashflow

loan

size

3313

48Num

berof

colla

teralloa

ns34

6Collateralloa

nsize

7125

113

2324

14Num

berof

firm

s13

118

020

04 Rev

enue

1,37

028

64,02

880

424

52,32

6Inve

stmen

t67

1320

559

1314

7Fixedassets

321

7287

519

758

487

Net

profi

t14

734

372

302

3520

03Total

employ

men

t48

3144

4929

48Lev

erag

e12

220

177

25Num

berof

firm

swith

indep

enden

taud

itors

7381

Num

berof

cash

flow

loan

s16

6Cashflow

loan

size

4321

68Num

berof

colla

teralloa

ns57

27Collateralloa

nsize

5920

109

4719

77Num

berof

firm

s15

727

3

Notes:T

reatmen

tgrou

pconsists

ofMSM

Esthat

hadreceived

aloan

from

oneof

eigh

tEBRD-spo

nsored

MSM

Elend

ingprojects

in2002.C

ontrol

grou

pconsists

ofMSM

Esthat

hadne

verreceived

anEBRD

projectloan

.Finan

cial

dataareexpressedin

thou

sand

sof

USD

s.Figu

resaread

justed

toprod

ucer

prices.F

orallc

ountries,the

coun

trysp

ecificprod

ucer

priceinde

xisus

edexcept

forRus

siaforwhich

theNizhn

yNov

gorodregion

alprod

ucer

priceindex

isus

ed.T

otal

employ

men

tisafulltim

eeq

uiva

lent

offull-tim

e,pa

rt-tim

ean

dtempo

rary

employ

ees.Lev

erag

eisdefi

nedas

ratio

ofdeb

ttodeb

tpluseq

uity.

Appen

dix

1.(C

ontinued

)



Appendix

2.Financialinterm

ediariesprovidingcashflow

loans

Finan

cial

interm

e-diaries

Cou

ntry

Yea

rprogra-

mme

started

Total

number

ofM

SM

Eloan

sex

tended

aten

d20

02since

beg

inningof

program

me

Total

volume

ofM

SM

Eloan

sex

tended

aten

d20

02since

beg

inningof

program

me

(million

Euros)

Loa

nsin

arrearsov

er30

day

sas

%of

outstanding

MSM

Eloan

portfolio

Ave

rage

MSM

Eloan

maturity

forloan

sex

tended

in20

02(number

ofmon

ths)

Number

ofbranch

esat

end20

02

Number

ofloan

officers

aten

d20

02

Ave

rage

borrower

size

interm

sof

number

ofem

ploye

es

Procredit

Ban

kBulga

ria

Bulga

ria

2002

5,91

939

.60.0

1610

3828

Procredit

Ban

kGeo

rgia

Geo

rgia

1999

9,97

347

3.1

1816

124

KMB

Rus

sia

1998

28,819

220.4

1.0

1866

6111

Procredit

Ban

kUkraine

Ukraine

2001

7,51

842

0.5

1618

7012

Heb

ros

Ban

kBulga

ria

2001

284

7.7

0.0

1835

4325

Tbiluniv-

ersal

Ban

k

Geo

rgia

2000

300

1.4

0.0

182

95

NBD

Rus

sia

1998

1,48

410

.11.0

na10

22na

Private

Ban

kUkraine

1999

4,41

124

.40.6

1145

130

11



Cash flow vs. collateralbased credit - CORE

Documents